1. Field of the Invention
The present invention relates to method and apparatus of image processing, and, more specifically, to method and apparatus of image processing capable of performing gradation reducing process in which number of gradations of image data is reduced, by using threshold values.
2. Description of the Related Art
Handling of images in digital manner is currently dominant in the field of image processing. It is often the case that for displaying or outputting digital image, it becomes necessary to display gradations of the image in smaller number of gradation levels, because of restrictions imposed by characteristics of an output device and the like. From the early stages of development, various methods of digital half toning image processing such as binarization, in which gradations are reproduced solely by white and black dots as a pseudo halftone processing, has been studied.
Various methods including ordered dither method and error diffusion method, which are still utilized at present, as well as descendents of these methods have been developed and improved from 1960's. Further, as the hardware of computation has been developed recently, a method of directly performing optimal search for pixel arrangement, such as the method of cost minimization, has been developed.
These methods of half toning have respective advantages and disadvantages in accordance with the objects of use, and therefore various problems and solutions for respective methods have been studied. For example, the ordered dither method is simple and easy to use, while reproduced image quality is not very good. Though load of computation is heavier in the error diffusion method than the dither method, image quality is better.
In the method of directly performing optimal search such as the method of cost minimization, various optimization methods such as neural network, genetic algorithm and simulated annealing are utilized. Adoption of such a method facilitates incorporation of a visual model or an output device model into the process, enlarging degree of freedom in the processing. On the other hand, as the optimal state is searched through repetitive operations, load becomes formidable.
The problems change along with the development of technology. The problem of formidable load experienced when the method of directly performing optimal search is used may be solved by the development of hardware defining the speed of calculation. From the viewpoint of promoting wide spread use of simple and high quality output devices, however, simpler calculation process is desired.
Further, there are the problem of trade off between resolution and gradation common to all the methods. This problem may possibly be solved by increased output gradation levels or improved resolution characteristic of the output device itself. It is expected, however, that there will be increased occasions where characters are processed as images, and such processing should desirably be done in the simplest manner possible.
Conventionally, methods of improving image processing have been studied, including a method in which an image region of which gradation is of importance and an image region of which resolution is of importance are determined and the method of processing is changed in accordance with the result of determination for respective regions, and a method in which a plurality of processing methods are combined. These methods are hardly said to be simple methods, as a new process of region determination, for example, must be developed and added to execute such methods. Considering balance with the hardware (output device), it is desirable that satisfactory resolution and gradation are both attained through such a method that is comparable to the error diffusion method.
FIG. 49 is a block diagram showing a configuration of a conventional image processing apparatus executing the error diffusion method.
Referring to the figure, the image processing apparatus includes: an input unit 501 receiving as an input a pixel value of one pixel of a multi-value image; a subtractor 503 subtracting diffused error from the input pixel value; an output unit 505 outputting, as a corrected pixel value, an output from subtractor 503; thresholding unit 507 performing thresholding on the output of output unit 505 to provide binary data; an output unit 509 outputting, as pixel data, the output of thresholding unit 507; a subtractor 511 subtracting the output of output unit 505 from the output of thresholding unit 507; and an error memory 513 for diffusing the output result from subtractor 511 to pixels around a pixel which is the object of processing (pixel of interest).
The image formed through error diffusion method has a particular texture. The texture is not very noticeable visually, as it has blue noise characteristic. A method of setting dither pattern to attain the blue noise characteristic in a simple manner has been studied for the dither method as well. In the error diffusion method, however, dot patterns are adaptively generated with respect to the input image, and therefore characteristic of the input image is better reflected than the dither method.
In this point, the error diffusion method is superior in image quality to dither method. The error diffusion method, however, has its particular noise. Namely, there occurs a phenomenon in which variation in texture at a region where gradation changes moderately results in an apparent border line where there is no border (texture shift), or a phenomenon in which white or black dots tend to appear in a line at a region where the gradation is close to black or white.
Various methods for improving have been developed to prevent these phenomena, including modulation of weight coefficient and threshold value for error diffusion. As to resolution, though inherent edge enhancement characteristic has been pointed out, it is not sufficient.
Further, from the nature of its algorithm, the error diffusion method functions to reproduce pixel values of the input image in averaging manner. More specifically, the method functions to reproduce local 0th order component of the image. Accordingly, the error diffusion method has been improved to enhance components of 1st and higher order.